Overview of solar position data and its relationship to weather and training data¶
Does the solar azimuth data align with solar radiation data?¶
This means that the day / night cycles align, as should be the case for data from the same location.
In (latitude, longitude), the coordinates are
Solar data: (58.60, 25.01)
Close weather station: (58.5, 25.2)
Zooming in on the plot, these actually seem to line up very well. As a qualitative check, we can filter by times when the the solar elevation is negative
Ok, it's not perfect, but the deviations are relatively small.
Does this hold for all weather station locations?¶
Looking through the different locations, a couple of things stand out
- This dataset is good enough for most locations, except two that have large night-time illumination: (57.6, 23.2), and (57.6, 24.2)
- both datapoints in the Gulf of Riga
- Larger longitudes have an error peak in July, while smaller latitudes have more concentrated errors in May and April
Does the solar data at this specific locations (57.6, 23.3) match the weather data better?¶
This should tell us how different the solar elevation data at the two locations is. The answer is not very. This pattern looks quite similar to the 'normal' timecourse of errors that we saw above, which probably results from the slight differences in location. It does not explain the massive deviations that we see.
Is the timezone off for this weather data?¶
This is off by more than just a shift of hours. The day and night lengths seem totally wrong for the season!
Indeed this datapoint seems quite off. The control looks like:
Both anamolous spots share this pattern
Answer: No.¶
This data is very strange. It really seems like it's not correctly recorded or not correctly associated to the time / date in question. It think that the best thing to do initially is to simply filter out the weather data at these particular locations
Filtering out the these datapoints, how does exact location data compare to a single location?¶
Total night radiation, same location 80838.0
Total night radiation, single location 86045.0
It is indeed slightly better to . I'm not sure that thats worth much, but the much of the improvement appears to be in the most extreme data points, so it may be worth something.
Comparing the solar day-night cycle to production cycles¶
This show that the solar data is well aligned with the production data, as we might hope.
More generally, because the daylight-vs-time relationship is a reflection of both latitude (in terms of fluctuations in day-length) and longitude (in terms of sunset times), it can give us a hint of the location of the production units. Of course, the local terrain will impact these things as well, so it is unlikely that we will be able to achieve resolution better than county-level.
However, we might see an improvement in how well these match by comparing to the county-level solar data. I might explore this later.